Ocean Environment,floating Body Response Prediction And Abnormal Diagnosis Of Underwater Structures Based On Deep Learning Method

As one of the important production equipment for offshore oil and gas development,the Deep-water floating platform is widely used in the Bohai Sea and the South China Sea.Semi-submersible Floating Production System(Semi-FPS),as a kind of deep-water floating platform,due to its unique structural characteristics,is widely used in deep-sea areas without being limited by the operating water depth.However,under the combined action of complex environment loads such as wind,waves,and currents,its motion responses present strong nonlinear and non-stationary characteristics,which will affect the safety service of the platform.Real-time monitoring and evaluation of the health status of the Semi-FPS is an important prerequisite to ensure its safety service.The current on-site monitoring methods and technologies for the platform floating bodies have become mature,but there are deficiencies in the monitoring methods for underwater mooring systems.The intelligent real-time monitoring system of the underwater mooring system is the guarantee for real-time diagnosis of its health status.In addition,the research about environmental load characterization and prediction,floating body motion response analysis and early-warning assessment based on the real data obtained by the existing monitoring system can accurately reflect the real health status of the platform.To the end,the thesis has carried out research on the ocean environment,floating body response prediction and abnormal diagnosis of underwater structure based on the years of measured information accumulated by the author’s research group.The main researches and results are obtained as follows:(1)Data-driven environment forecast method with multi-step training set extensionTo solve the problems of the environment load predictions with complex and changeable characteristics,a novel data-driven environment forecast method with a multi-step training set extension is proposed.Firstly,based on the measured wind data of the South China Sea,the wind speed change pattern is extracted by using deep learning methods and a short-term wind speed forecast model is constructed for the South China Sea.The prediction error is only0.97m/s.Secondly,the correlation analysis of wind speeds and wave heights is carried out and different time intervals are selected to extract wave height and wind speed extreme values.Finally,a wave height forecast model with multi-step training set extension is established based on the Long-Short-Term Memory(LSTM)method.Simulated results show that the 6h prediction error is reduced by 45.83% compared with the one-step model and the 12 h prediction error decreases by 52.83%.(2)Motion prediction and early warning assessment of semi-FPS considering a physical processThe research on motion prediction and early warning assessment of Semi FPS is carried out considering physical process.Firstly,based on fractal theory and statistical analysis methods,the measured environment information is analyzed for dimensionality reduction,and a physical process-based method for extracting mixed features of ocean environment loads is proposed.Secondly,based on the LSTM network,a short-term prediction method of the platform’s motion responses is established.The prediction error is only 0.07°.In addition,based on the measured information of the Six-Degrees-of-Freedom(6 Do Fs),the early warning index is given and the early warning assessment model is established.The prediction results show that the accuracy reaches 86.30%.What’s more,for the safety service of platform workers,the personnel operation risk index is calculated based on ISO6897-1984(E).The relationship model is established between environment feathers and the operation risk index based on Sparse Auto Encoder(SAE)and the evaluation method is proposed which can guide platform operation and safety service.(3)Acoustic transmission strategy for underwater inclination signal acquisition in a multi-point mooring system based on compressive sensingDue to the obstacles to underwater acoustic transmission such as narrow bandwidth and data loss,a novel acoustic transmission strategy for underwater inclination signal acquisition based on compressive sensing is proposed.Firstly,time-frequency analysis is carried out on the measured inclination data and a calculation method of the top tension force of the anchor chain is established.Then,to address problems like narrow bandwidth and easy data loss,a time-domain signal compressing-reconstruction algorithm is constructed and the acoustic transmission system is proposed.Finally,the effects of compression ratios and loss rates on the reconstruction accuracy are analyzed to demonstrate the accuracy,feasibility,and reliability of the proposed system,which can provide a guarantee for real-time assessment of the health status of the underwater mooring system.(4)Abnormal diagnosis and positioning method of underwater mooring system combining LSTM and PCAAiming at the difficulty of damage diagnosis and identification of the underwater mooring system,an abnormal diagnosis framework of underwater mooring system is proposed combining LSTM and PCA(Principal Component Analysis).Specifically,based on the real design parameters of a semi-submersible platform,a high-fidelity hydrodynamic model is established.Secondly,different damage degrees are chosen and the platform’s motion responses at different positions are analyzed.Furthermore,based on the LSTM network and normal state data,a joint prediction model of the platform’s 6Do Fs motion is established considering the marine environment loads.The PCA method is tailored and employed to build an abnormal diagnosis model combing with the prediction residual sequences and the Upper Confident Level(UCL)is given.In addition,the abnormal database of the underwater mooring system is built based on the simulation model.The damage positioning model of the underwater mooring system is established by using the Deep Sparse Auto Encoder(DSAE)method,which can guide the real-time diagnosis of the underwater mooring system.The positioning accuracy reaches 100% when the damage degree is over5%.